Modelling with SAGE: lessons and future plans
description
Transcript of Modelling with SAGE: lessons and future plans
1
Modelling with SAGE: lessons and future plans
Jane Falkingham & Maria Evandrou
ESRC Centre for Population Change
University of Southampton
BSPS Annual Conference, University of Sussex
11th September, 2009
2
Outline
• Introduction
• Overview of the SAGE microsimulation model
• Challenges and lessons
• The Future
3
Introduction
• ESRC Research Group ‘Simulating social policy in an Ageing Society’ (SAGE) funded 1999-2005; originally based at LSE and KCL (Falkingham, Evandrou, Rake & Johnson)
• Main aim: “to carry out research on the future of social policy within an ageing society that explicitly recognises the diversity of life course experience”– Substantive research on the life course– Development of a dynamic microsimulation model– Exploration of alternative policy options
4
Simulating life course trajectories to 2050: the SAGE Model
• Project likely future socio-economic characteristics of older population– Family circumstances– Health & dependency– Financial resources
• Project future demand for welfare benefits & services among older people
• Assesses impact of social policy reform scenarios
5
Overview of characteristics of the SAGE Model
• Base population: 0.1% of GB population = 53,985 individuals
• Partially closed (internal marriage market)• Transitions – both deterministic and stochastic • Discrete time (rather than continuous)• Time based processing (rather than event
based)• C++• Efficiency in processing → quick run times
6
Contents of the SAGE Model
• Demographic – Mortality– Fertility– Partnership formation– Partnership dissolution
• Health– Limiting long-term illness– Disability
• Employment– Paid work– Unpaid work (informal care)
• Earnings• Pensions
– Public– Private
• Other Social security transfers– Pension Credit, disability living allowance, attendance allowance
7
SAGE Model Base population
• 10% sample of 1991 Household SARs and 5% of institutional residents from 2% Individual SARs
plus • Additional characteristics• Data matching / Donor imputation
– Duration of partnership (BHPS)– Missing labour market characteristics– Pension contribution & caring histories (FWLS)
• Regression imputation– Aligning limiting long-term illness (QLFS)
8
A B C
donordonorrecipientrecipient
SARsSARs BHPSBHPS
Duration of Duration of partnershippartnership
Matching variablesMatching variables
Donor Imputation: eg duration of partnership
9
SAGE Model Transition Probabilities
• Mortality ONS LS, GAD
• Fertility & Partnership BHPS, GHS• Health QLFS• Disability BHPS• Employment QLFS • Earnings BHPS• Pension scheme membership FRS• DLA and AA BHPS
10
SIMULATION
INPUT (BASE) DATA
INPUT (BASE) DATA
OUTPUT DATAOUTPUT DATA
SAGE Model programming structure
POPULATION EVENT LIST
LOG FILELOG FILE SCRIPT FILESCRIPT FILE
CONSOLE
1991
1993
1995
1997
1999
11
Challenges
• Technical– Validation– Alignment (fig 1a, 1b)
• Operational– Timeliness– Maintenance– Sustainability
12
Fig 1a: Proportion in employment by birth cohort Men, 1995- 2050
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Age
1930-40
1940-50
1950-60
1960-70
1970-80
1980-90
1990-00
2000-10
2010-20
2020-30
Source: SAGEMOD
13
Fig 1b: Proportion in employment by birth cohort Men, 1995- 2050 (aligned to HM treasury forecasts)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Age
1930-40
1940-50
1950-60
1960-70
1970-80
1980-90
1990-00
2000-10
2010-20
2020-30
Source: SAGEMOD
14
Lessons
• Microsimulation models are resource hungry– Data– Human resources (DWP MDU c.20; SAGE
1fte programmer and 1fte analyst)
• Ideal team involves range of skills– At a minimum need demographer, economist,
statistician/ operational researcher, social policy analyst and computer scientist
15
Lessons
• Time spend in efficient programming reaped rewards in short run times
• Minimising ‘embedded’ parameters maximising ‘what if’ scenarios
• Desktop user model increases flexibility• Sharing expertise across modelling groups
(PENSIM, SESIM, MOSART, DYNACAN, DYNAMOD)
But• No quick fix, every model and every social
system different
16
Future plans
• Development of dynamic multi-state population model within CPC (ESRC)
• Collaboration with University of Southampton colleagues in Centre for Operational Research, Management Science and Information Systems (CORMSIS) and Institute for Complex Systems Simulation (ICSS) on updating and extending SAGE model (EPSRC)
• Incorporation of uncertainty and expert opinion through Participative Modelling
17
Selected publications
M. Evandrou and J. Falkingham (2007) ‘Demographic Change, Health and Health-Risk Behaviour across cohorts in Britain: Implications for Policy Modelling’ pp. 59-80 in A. Gupta and A. Harding (eds.), Modelling Our Future: Population Ageing, Health and Aged Care, International Symposia in Economic Theory and Econometrics, 16, Elsevier.
M. Evandrou, J. Falkingham, P. Johnson, A. Scott and A. Zaidi (2007) ‘The SAGE Model: A Dynamic Microsimulation Population Model for Britain’ pp. 443-446 in A. Gupta and A. Harding (eds.), Modelling Our Future: Population Ageing, Health and Aged Care, International Symposia in Economic Theory and Econometrics, 16, Elsevier.
A. Zaidi, M. Evandrou, J. Falkingham, P. Johnson and A. Scott (2009) ‘Employment Transitions and Earnings Dynamics in the SAGE Model’ pp. 351-379 in Zaidi, A. and Marin, B. (eds) New Frontiers in Microsimulation Modelling Aldershot: Ashgate.